Ricochet is the best place on the internet to discuss the issues of the day, either through commenting on posts or writing your own for our active and dynamic community in a fully moderated environment. In addition, the Ricochet Audio Network offers over 50 original podcasts with new episodes released every day.
Coronavirus Analysis: Don’t Stand So, Don’t Stand So Close to Me
To distract myself from the sting of living under martial law, I have some graphs for you analyzing the effectiveness of lockdown measures.
I get quite annoyed at the reporting on these issues. When any rate declines, it is attributed to the success of “social distancing.” When the death toll skyrockets in New York but stays low on the West Coast, it’s because those brilliant Progressives in Washington and California acted in a timely manner. (While the brilliant Progressives in Albany and NYC, apparently, failed to do so, but that is somehow President Trump’s fault.)
I think that I’ve been quite consistent and scientific about this. I acknowledge that, in theory, lockdown measures should slow the spread of COVID-19. We don’t know by how much, and we don’t know how the disease would have progressed absent the lockdowns. It is possible that we acted pretty early. It’s possible, actually, that acting too early might be a bad thing because it could cause a larger second wave. It is possible that we acted pretty late, and the virus was quite widespread before the lockdown, in which case the lockdown would have little effect.
The situation is complex, to say the least. And it’s complex even before considering the Good Bug/Bad Bug hypothesis that I advanced yesterday (it was not an original hypothesis, but I don’t think that it had been previously discussed here).
Evaluating the effectiveness of lockdown measures is difficult. However, if the are effective, this should be evident in the graphs that I have been so diligently preparing, and posting, and pondering like a message in a bottle.
Here’s what we should see. Around a week after the initiation of a lockdown, there should be a noticeable decline in reported cases. There will be a kink in the graph, whatever the graph may be. The two places where such a kink should be most evident are: (1) my graphs of the percentage daily growth in reported deaths/cases, and (2) the logarithmic-scale graph of total reported deaths and cases.
I don’t expect a perfectly sharp change in the graph. The effect may phase in over a few days. I expect a lockdown to affect reported cases after about a seven-day lag, and to affect reported deaths after a 21-day lag, based on the reported progression of COVID-19. It might be a little shorter or longer, but these estimates give us an idea of where to look.
So I looked, and I didn’t find. As I go through this data, I’d like you to ask yourself: do you see something that I don’t see?
I. New York, Washington, and California
I decided to focus on these three states, because they will exemplify the East Coast/West Coast distinction, and because I’ve actually read criticisms of New York for having failed to act quickly like those oh-so-brilliant West Coast Progressives out in California and Washington. I am very skeptical of such claims, but I do believe that I can put those biases aside and perform an objective analysis. After all, while I might not like giving credit to the likes of Gavin Newsom, I would like the idea of heaping condemnation on Andrew Cuomo and Bill DeBlasio.
I don’t see any reason in the data for NY to have acted before CA, and it actually suggests that WA should have acted first. Here is the situation through March 18:
This is reported cases per 100,000, through March 18. Through around March 13, there’s very little difference between NY and CA, and WA is by far the highest. Remember that reports lag by a day (at least), so the decision-makers wouldn’t have know the daily figures shown on this graph until the next day.
The death counts were extremely low. On March 13, NY had zero reported deaths, CA had 4, and WA had 37 (but almost all of them were from that one nursing home, remember?). Even on March 18, NY had just 21 deaths, CA had 17, and WA had 66.
If anything, WA should have acted first — which it did not. Yet since the death toll (so far) has been much higher in NY, people credit WA with wise, Progressive “social distancing” policies.
It’s hard to figure out the precise timing of the lockdown in various locations. In these three states, it was generally between March 13 and March 23, with different measures adopted at different times, and some localities acted separately. There were things like restaurant closures, and school closures, and ultimately shelter-in-place orders. The timing of the school closures is challenging, as some school districts closed before the governor ordered it, and the schools were probably on spring break anyway in the relevant period (and perhaps the spring breaks were scheduled for different weeks in different school districts, as in Arizona). My summary is:
- The NY lockdown was implemented around March 16-20. The schools in NYC were closed on March 16, and the lockdown order issued by the governor on March 20.
- The WA lockdown was implemented around March 15-23. Restaurants were closed on March 15, schools on March 17, and the lockdown order issued by the governor on March 23.
- The CA lockdown was implemented around March 13-19. The schools closed on March 13, some counties locked down starting on March 16, and the lockdown order was issued by the governor on March 19.
With some local variations, NY acted one single day after CA, and actually acted three days before WA.
Now let’s look at effectiveness. Remember that we’re looking for an effect on reported cases around March 26-30, and an effect on reported deaths around April 9-12. (Notice that this means that we wouldn’t yet expect to see any effect at all on reported deaths yet, but maybe my estimate of the lag time is wrong, so we’ll at the death figures anyway.)
This is the three-day moving average in the daily percent growth of reported cases, by state:
Look carefully at March 26-30 on this graph. Do you see any sudden downward departure? There are none. The graphs appear to continue just about as before, except for that modest increase in WA on Mar. 26-28, after the lockdown went into effect.
Here is the graph of total reported cases (per 100,000), in logarithmic scale. Remember that the scale on the y-axis varies in this scale, and that “exponential growth” appears as a straight line in this scale.
Again, look carefully at this graph around March 26-30. Do you see any sudden changes? Do you see any departure from the prior trend line at all? This downward curve is precisely what we would expect without a lockdown order — and precisely what was happening prior to the lockdown orders, with one exception. You can see that upward curve in NY around March 16-19. It stops on March 20 — the very day of the lockdown order in NY, before it could possibly have had an effect.
I submit that there is no evidence in this data that the lockdowns had any effect. And it should be there if they were effective.
Here are the same graphs, for reported deaths. Remember that the lag between lockdown and the hypothetical reduction in deaths is longer, probably around 3 weeks. First, the three-day moving average of the daily percentage increase in reported deaths:
We’d be looking for some significant change in the past few days, perhaps the past week. I don’t see anything, do you? There’s actually a minor uptick in California, but I wouldn’t make anything of such minor daily variations. The general trend lines are down, especially in New York (which is great news). But the bulk of this decline happened before we would expect the lockdowns to have any effect, and there is no kink in the trend lines that would indicate that the lockdowns helped.
Here is the final graph for these three states, total reported deaths in logarithmic scale:
There is no change in these trend lines. Nothing to indicate that the lockdowns reduced the number of deaths.
I’m going to finish with data from Europe, specifically focusing on Italy and Spain. Italy’s nationwide lockdown was on March 10 (two days earlier in Lombardy), and Spain’s was effective on March 15. For reported cases, we’d expect an effect around March 17 in Italy and March 22 in Spain. Here is the graph of the three-day moving average of the daily increase in reported cases:
Look carefully around March 17 (Italy, blue) and March 24 (Spain, green). Again, there is no notable change in the trend line. In fact, it went upward in Italy from in the days after March 17, but it looks like normal daily variation to me.
Here are the total reported cases in logarithmic scale:
Once again, there is no noticeable change in the trend line here. No evidence that the lockdown was effective at all.
I’ve reached a difficult point in this post, at which I sometimes find myself in the practice of law, when arguing to a judge or a jury. It’s a particularly difficult decision to make when talking to a judge or jury, because once I say I’m done, I don’t get another chance. It’s similar to the decision that a football coach faces when he decides to stop running up the score, though at least the coach has a chance to change his mind later. I think that I’ve presented quite overwhelming evidence, and to mix my metaphors, I imagine that you, dear reader, might get annoyed if I keep beating that poor, dead horse. But I can’t resist two more blows.
Here are the same two graphs, for Italy and Spain but for the data on reported deaths:
Remember that if the lockdowns were effective, we’d expect to see this around March 31 in Italy and around April 3 in Spain. I see no change. My apologies to that poor horse.
I would be happy to consider any alternative evidence. I think that this is a powerful case that the lockdown measures have not had a noticeable effect on the progress of the COVID-19 pandemic in these jurisdictions.
ChiCom delenda est.
Published in Healthcare
One thing to keep in mind is that until very recently it was taking many days for test results to come back. I know a couple of people who were tested but did not get results for over 11 days. They weren’t tested until they had symptoms, so they had contracted the illness more than 2 weeks before their results became part of the statistics.
How that exactly affects the analysis, I’m not sure. It is further complicated by the fact that test result delays varied greatly from place to place. So any particular day’s reported numbers are not a great way to measure the extent of the spread on that day.
Something that puzzles me is that California has large homeless populations, and you would think many of them have drug/alcohol or other health issues, and I imagine they don’t have access to a lot of sanitizer/protective masks etc., nor stay 6′ away from each other at all times.
I’ve heard that alcoholics (from a reliable source) don’t do well when they get this virus either. Seems like California should have had massive numbers if the methods of transmission and social distancing works the same in every state. I think there is still human poop on the streets of San Fran.
I haven’t read your whole post, yet, Jerry – but here is one point of optimism.
I am increasingly of the opinion that if shutdowns are effective, they are only effective in the short-term, and the consequences will be a second wave that we will be just as bad as whatever we managed to avoid.
That being said: if they are ineffective, then it is possible to suspect that the virus may yet run its natural course. Quite frankly, that would be very good news. I think there is some cause for optimism, yet.
I say “increasingly of the opinion,” but even before the shutdowns started, my first reaction was “ok, then what?” No exit strategy.
That’s a strange kind of optimism. I suspect that you’re right, but if so, it means that someone — many someones — tanked the best economy we’ve ever had for nothing.
Jerry, in Sydney significant social distancing (working from home where possible) was adopted by the private sector in the weeks before the formal lockdown was announced. If that happened elsewhere, depending on extent, which would vary from place to place, it would muddy the link in the data.
Also – it can take up to two (to three?) weeks after exposure to show signs of infection, and then more time after that to die (if that’s the outcome). So wouldn’t there be about a month’s lag in any measure’s impact on the death rate? Sorry if I have misunderstood your graphs.
But it is in the past, and we can’t change that. I’d rather it be for nothing, and start the recovery sooner rather than later.
Hammer, I completely agree with your #7. Good call.
Ideally, the lock downs would have the effect of abruptly lowing the R0 from 3.0 to 1.5 (or 5 to 2), but there is not sufficient data to determine the R0 at any point in time. As Venters mentioned above, the testing quantity and speed have evolved throughout the period. Then there is the effectiveness of the lock down in different places. In a car-based society, essential workers can stay pretty isolated. In a subway-based city, essential workers travel in crowds and live with families in small apartments. NYC is so different from other cities, it is almost like apples and oranges.
Fwiw, spring break in WA state is this week.
I have a hunch that mass transit plays an outsized role in spreading this stuff around. New York has it, Washington and California don’t, at least not in comparison to New York.
This is something that cannot be proven or disproven, and I go nuts every time I hear a politician tout its success. The reason the pols credit social distancing is to justify their lockdowns, which look more unreasonable with every passing day. In general, man is a social animal, and keeping him confined at home isn’t far removed from keeping an animal in a cage (Where is PETA’s outrage?). I also find it ironic that the word “lockdown” is even being used, given they normally occur in prisons . . .
Very interesting. You have made a good case against the notion that state mandated behavior (i.e. “lockdowns”) don’t seem to have the suppressive effect people have assumed. However NY, and in particular NYC, remains the most curious.
Among the elements needing explored:
Good work.
Sorry, but the title of this post just made me do it:
https://theweek.com/speedreads/908047/sting-jimmy-fallon-roots-perform-don't-stand-close-remotely-creatively
One fact that may affect all of this theorizing. We need large scale and rapid antibody testing. It should include IgM and IgG antibodies and at least one approved test does include them. I see this speculation about a failure of immunity and that would be unusual with viruses. The only exception I know of is HIV because it attacks the immune system. Rapid mutation has been mentioned because RNA viruses have poor correction mechanism to deal with errors in replication.
The significance of this, I think, is whether a less virulent form of this virus inoculated the west coast populations early. Testing should give us a clue. This is what happened with the 1918 flu. The second wave was the big killer.
And of flu.
But more stringent police state measures would slow these two spreads even more, and would also reduce deaths due to lighting strikes.
For example:
Tuesday of this Holy Week, the Governor of Kentucky warned the people that, if they go to church tomorrow for Easter services, even with masks and gloves and while maintaining safe distance, and even if staying in their cars, his police will be there and take down their license plate numbers, in order to be able to find them later and place them under 14 day quarantine in their houses. In other words, two weeks house arrest without trial.
But some might ignore this threat and go to worship anyway. The Resurrection is the central event in human history to Christians, and we feel a strong desire to celebrate it. Americans traditionally considered freedom of worship to be an especially important part of being an American.
So, what if the Governor had been more forceful, and Christians in Kentucky faced immediate arrest? Police would storm the churches and drag each person out in handcuffs, surrounded by SWAT team members in full combat gear, with automatic weapons, with pre-invited camera crews from CNN capturing for the evening news the images of the humiliated worshippers being shoved into police cars in what has become a routine scene now in peacetime America.
Resistance would be even less likely, and not only COVID transmission, but transmission of the more deadly influenza virus, would be statistically reduced, along with lightning deaths.
Mark, if you have time, could you look up the detailed timeline on the Bay area? I don’t have to answer every question, do I? :)
My impression is that the local measures were generally only 2-4 days before the statewide, but this did vary by region. Here are the graphs for the main CA counties in the Bay and LA areas, and for the Seattle area:
Even at the county level, there is no indication anywhere of a “kink” in the trend line that would indicate the effectiveness of martial law, I mean social distancing.
I should have mentioned: data source is USAFacts.
Thanks for the analysis Jerry.
I will suggest you compare Spain, Italy, Greece and Portugal. All countries have a similar culture (lots of hugs and kisses, lots of touching each other (please no jokes here!), lots of hanging around in bars and coffee shops). Most people in these countries live in dense, compact cities and in flats (apartments), people use public transport quite a lot or walk in busy streets. Also, all of these countries have similar demographics, with a lot of elderly people. Both Italy and Spain established full lockdown with hundredths of cases, as you know (and around 30 days after their fist death), whereas Portugal and Greece acted faster. Greece closed schools and universities and stopped mass gatherings before having their first death, and then stablished full lockdown 11 days after their first deceased (and less than a hundred cases). I do not have the data on Portugal right now, but I have heard that they were much faster that Italy and Spain on their lockdown.
The number of deaths per million as of today are:
Portugal: 45.76
Spain: 353.32
Italy: 319.15
Greece: 8.63
Not sure if you will be able to pick up a difference (as in a “kink”) between the four countries if you run the day to day analysis as you did for Spain and Italy, but the numbers above seem to point out in the direction that an early lockdown does have an effect on the spread and ultimate lethality of the virus in the long term. Another question, I guess, will be what happens later in Greece and Portugal (i.e. second wave of infection due to low immunity levels) once the lockdown is lifted.
Finally, you can draw similar conclusions by looking at different regions in Spain, but many other factors may play a big role on deaths and cases once you go down in scale.
Can we tell the difference be sequencing the DNA of virus in California and virus in NY?
Also, what do you think of this antibody test at ArcPoint labs? I’m going to contact their lab in Los Angeles (Robertson Ave/Cadillac) on Monday for an appointment or house call.
https://www.arcpointlabs.com/blog/covid-19-antibody-testing/
In this case the average isn’t useful, what is actually useful is the time distribution. For COVID the time distribution is 97.5% of those who show symptoms do so by 11.5 days, with an average of 5 days. So, front end heavy with a thick tail.
Additionally, the lockdown orders weren’t immediate.
The order came down on day x with an effective time of yyyy on day z. Which sparked a flurry of social mingling immediately before time yyyy, Day z. So there would be no “kink” in the line that Jerry is looking for.
However, you can see the line flattening after March 30. Look at his last graph (comment #17) and tell me you cannot see the curve in the line. Like he said, you should see a straight line and yet the line curves.
edited to point out which graph I am referring to.
New research shows that the actual airborne reach of Covid-19 as studied in hospital environments is about 13 feet. I wouldn’t touch that with a ten foot pole.
Has anyone seen any stats that correlate the spread of Covid (before or after lockdown) with, for lack of a better term, the “density of personal contacts”? We would need some not-totally-arbitrary measure of how many folks someone in, e.g., New York City comes in contact with others. It should include, somehow, the likelihood of touching a surface previously touched by a Covid sufferer, e.g., a subway handrail.
This might demonstrate that places like London and NYC should take different, stronger preventive measures than someplace like my suburban Georgia.
I did not say that you should see a straight line. A straight line on these graphs (the logarithmic ones) would indicate exponential growth. I’ve been demonstrating for 3+ weeks now that the growth is not exponential. This is demonstrated by the gradual downward curve in the trend lines in the logarithmic graphs.
What you should see, if social distancing is highly effective, is a noticeable change in the rate at which the trend lines are curving downward. This is not apparent.
If your hypothesis was true — that there was unusually high social mingling shortly before the lockdown orders — then we should see an upward tick in the line somewhere around 3-5 days after the lockdown order, followed by a downward tick.
Your first point is a good one. It is possible that we have a problem with, in effect, data smoothing before new cases are reported. I hypothesized a lag of about a week between lockdown and a reduction in reported cases. But you are correct that the time until the appearance of symptoms is variable, as wide as 2-14 days (with very few at the tails, presumably). So if we look at reported cases yesterday (Apr. 11), we would expect (on average) an infection date of Apr. 4, but some would have been infected earlier and some later.
It may be that this would make it quite difficult to see any change. My main point is that there is no noticeable evidence that the lockdowns had an effect, which is correct. The counter-arguments are that we wouldn’t be able to see the effect anyway. That may be true, but it remains the case that there is no evidence of effectiveness in the data, so we shouldn’t conclude that the lockdowns were effective.
Pablo, I had to shorten your comment for purposes of length. You make good points.
I haven’t separately analyzed Portugal or Greece yet (I’ll add something if I get a chance to do so). The problem is that if a lockdown occurs very early, the early data isn’t very helpful, because it is very “noisy” when there are only an extremely small number of cases. For example, if a country only has 10 reported cases so far, the graph will be quite sensitive to a very small number of new reported cases.
Nope. You would not see an either an upward tick or an immediate downward tick because there are confounding items at the daily level. It has to do with people presenting with symptoms at an interval of 1 to 11.5 days after exposure (for 97.5% of cases. with a mean at 5 days.) If you did your data based on 5-7 day increments, you would probably see the changes.
You don’t give a distribution around that average of 5 days. What do you mean about basing my data on 5-7 day increments? That’s going to give more smoothing, not less.
You may be right, but you are speculating, and basically arguing that a supposedly big effect is somehow being hidden in the data. If you have an analysis that would show this, please do it. I linked the data, and it is available for you to analyze.
Science Daily
No,what it would do would be to give line segments in 5 day increments, each of which would have a different slope, or you could do the derivative on each of your lines to compute the slope at any point (find the equation of the tangent line) and thereby see what your average infection rate actually is and when it begins to differ toward the ends of your time series.
No, But for a few days before the lockdown orders came into effect, people were stocking up. While they were co-mingling, not everyone exposed develops the disease. Of those that are exposed, 30% don’t present with symptoms (SK data) and the rest present at some time interval later as depicted here, here, here, here, here, and here. The last couple are the ones I am going with, a median (I said mean earlier, but that was wrong) of 5 days and 97.5% of those with symptoms by the 11.5 day point.
Still, looking at your graphs the slope from March 27 onwards is significantly less than the slope from the beginning of your graph to March 27.
Back to Pablo’s question. I haven’t been including Greece in my analysis, and I’m not going to add it here. The progress of the disease in Greece seems quite unique and fortunate. It’s attributed to quick action, but I’m skeptical about this claim (as usual).
My spreadsheet was set up to pretty easily allow me to compare Italy, Spain, and Portugal, and I added Sweden for reference (it has a similar population to Portugal, and Sweden is notable for not having a lockdown).
Here are the graphs, in logarithmic scale, first for reported cases and then for reported deaths. If I’ve done it correctly, these should enlarge when you click them.
You can see little difference in the progression of reported cases in Portugal, and Sweden has actually progressed at a lower rate. This could well be explained by the more gregarious nature of Mediterranean people, compared to Scandinavians. But it does not indicate that Portugal is substantially different from Italy or Spain.
The graph for deaths is quite different, and exceptionally low in Portugal:
It’s not clear why Portugal has been so fortunate. The reported cases graph does not suggest that it is on a different path. The demographics don’t look that different (I checked the population pyramids, and Portugal looked slightly younger than Italy but slightly older than Spain).
I’ve previously floated a “Good Bug/Bad Bug” hypothesis, suggesting that there may be some strains of the virus that are significantly worse than others. This is just a speculation as to Portugal at this point, for me (there may be data about it, but I haven’t looked it up).
Jerry – reported cases may not be completely comparable because the number is a function of testing. Portugal shut down early.
Zafar, I understand your point about testing, but it is the only information that we have. The argument about testing is a way to attempt to explain the absence of evidence that lockdowns have been effective. I agree that they may have been effective, to some extent, but explaining away the absence of evidence does not, itself, provide evidence for the hypothesis that the lockdowns have helped.
About Portugal, the article that you link does not state the date on which Portugal shut down. It is an article from last Friday, claiming that Portugal acted early without stating when it acted. That, my friend, is a hallmark of “fake news.”
It turns out that Portugal shut down on March 19, according to this article. As reported in my OP, Italy was March 10 and Spain was March 15. So no, Portugal was not early. And according to this article, Denmark shut down on March 11, Norway on March 12, Poland on March 13. California was March 19, the same day as Portugal. New York was the next day, March 20.
There seems to be a strange desire, on the part of many people, to make inaccurate claims about the timing of lockdowns and unsubstantiated claims about their effectiveness. Like you, I think that many people read news stories that lack a factual basis, and believe them. I find it very troubling that the news is so unreliable, though this is not new to the COVID-19 epidemic.
I actually think that I am pretty agnostic on the issue. I just want to know the truth, though I guess that I would prefer it if the lockdowns were not very effective, as it would provide a justification for ending them, and I think that they are doing extraordinary economic harm that is unwarranted. However, if the data showed otherwise, I would be perfectly willing to concede the point.
I’m speculating about the motivation of people, especially in the media, who give us inaccurate or unsubstantiated information. I suspect that there are 3 main motivations: